This paper is published as part of Faraday Discussions ...gold.cchem.berkeley.edu/Pubs/DC216.pdf ·...
Transcript of This paper is published as part of Faraday Discussions ...gold.cchem.berkeley.edu/Pubs/DC216.pdf ·...
This paper is published as part of Faraday Discussions
volume 141: Water – From Interfaces to the
Bulk
Introductory Lecture
Spiers Memorial Lecture Ions at aqueous interfaces
Pavel Jungwirth, Faraday Discuss., 2009
DOI: 10.1039/b816684f
Papers
The surface of neat water is basic
James K. Beattie, Alex M. Djerdjev and Gregory G.
Warr, Faraday Discuss., 2009
DOI: 10.1039/b805266b
Negative charges at the air/water interface and their consequences for aqueous wetting films containing surfactants
Katarzyna Hänni-Ciunel, Natascha Schelero and
Regine von Klitzing, Faraday Discuss., 2009
DOI: 10.1039/b809149h
Water-mediated ordering of nanoparticles in an electric field
Dusan Bratko, Christopher D. Daub and Alenka Luzar,
Faraday Discuss., 2009
DOI: 10.1039/b809135h
Ultrafast phase transitions in metastable water near liquid interfaces
Oliver Link, Esteban Vöhringer-Martinez, Eugen
Lugovoj, Yaxing Liu, Katrin Siefermann, Manfred
Faubel, Helmut Grubmüller, R. Benny Gerber, Yifat
Miller and Bernd Abel, Faraday Discuss., 2009
DOI: 10.1039/b811659h
Discussion
General discussion Faraday Discuss., 2009 DOI: 10.1039/b818382c
Papers
Hydration dynamics of purple membranes
Douglas J. Tobias, Neelanjana Sengupta and Mounir
Tarek, Faraday Discuss., 2009
DOI: 10.1039/b809371g
From shell to cell: neutron scattering studies of biological water dynamics and coupling to activity
A. Frölich, F. Gabel, M. Jasnin, U. Lehnert, D.
Oesterhelt, A. M. Stadler, M. Tehei, M. Weik, K. Wood
and G. Zaccai, Faraday Discuss., 2009
DOI: 10.1039/b805506h
Time scales of water dynamics at biological interfaces: peptides, proteins and cells
Johan Qvist, Erik Persson, Carlos Mattea and Bertil
Halle, Faraday Discuss., 2009
DOI: 10.1039/b806194g
Structure and dynamics of interfacial water in model lung surfactants
Avishek Ghosh, R. Kramer Campen, Maria Sovago
and Mischa Bonn, Faraday Discuss., 2009
DOI: 10.1039/b805858j
The terahertz dance of water with the proteins: the effect of protein flexibility on the dynamical hydration shell of ubiquitin
Benjamin Born, Seung Joong Kim, Simon Ebbinghaus,
Martin Gruebele and Martina Havenith, Faraday Discuss., 2009
DOI: 10.1039/b804734k
Discussion
General discussion Faraday Discuss., 2009 DOI: 10.1039/b818384h
Papers
Coarse-grained modeling of the interface between water and heterogeneous surfaces
Adam P. Willard and David Chandler, Faraday Discuss., 2009
DOI: 10.1039/b805786a
Water growth on metals and oxides: binding, dissociation and role of hydroxyl groups
M. Salmeron, H. Bluhm, M. Tatarkhanov, G. Ketteler,
T. K. Shimizu, A. Mugarza, Xingyi Deng, T. Herranz, S.
Yamamoto and A. Nilsson, Faraday Discuss., 2009
DOI: 10.1039/b806516k
Order and disorder in the wetting layer on Ru(0001) Mark Gallagher, Ahmed Omer, George R. Darling and
Andrew Hodgson, Faraday Discuss., 2009
DOI: 10.1039/b807809b
What ice can teach us about water interactions: a critical comparison of the performance of different water models
C. Vega, J. L. F. Abascal, M. M. Conde and J. L.
Aragones, Faraday Discuss., 2009
DOI: 10.1039/b805531a
On thin ice: surface order and disorder during pre-melting
C. L. Bishop, D. Pan, L. M. Liu, G. A. Tribello, A.
Michaelides, E. G. Wang and B. Slater, Faraday Discuss., 2009
DOI: 10.1039/b807377p
Reactivity of water–electron complexes on crystalline ice surfaces
Mathieu Bertin, Michael Meyer, Julia Stähler, Cornelius
Gahl, Martin Wolf and Uwe Bovensiepen, Faraday Discuss., 2009
DOI: 10.1039/b805198d
Discussion
General discussion Faraday Discuss., 2009 DOI: 10.1039/b818385f
Papers
Similarities between confined and supercooled water Maria Antonietta Ricci, Fabio Bruni and Alessia
Giuliani, Faraday Discuss., 2009
DOI: 10.1039/b805706k
Structural and mechanical properties of glassy water in nanoscale confinement Thomas G. Lombardo, Nicolás Giovambattista and
Pablo G. Debenedetti, Faraday Discuss., 2009
DOI: 10.1039/b805361h
Water nanodroplets confined in zeolite pores
François-Xavier Coudert, Fabien Cailliez, Rodolphe
Vuilleumier, Alain H. Fuchs and Anne Boutin, Faraday Discuss., 2009
DOI: 10.1039/b804992k
Dynamic properties of confined hydration layers
Susan Perkin, Ronit Goldberg, Liraz Chai, Nir Kampf
and Jacob Klein, Faraday Discuss., 2009
DOI: 10.1039/b805244a
Study of a nanoscale water cluster by atomic force microscopy
Manhee Lee, Baekman Sung, N. Hashemi and Wonho
Jhe, Faraday Discuss., 2009
DOI: 10.1039/b807740c
Water at an electrochemical interface—a simulation study Adam P. Willard, Stewart K. Reed, Paul A. Madden and David Chandler, Faraday Discuss., 2009 DOI: 10.1039/b805544k
Discussion
General discussion Faraday Discuss., 2009 DOI: 10.1039/b818386b
Concluding remarks
Concluding remarks Peter J. Feibelman, Faraday Discuss., 2009 DOI: 10.1039/b817311g
Coarse-grained modeling of the interfacebetween water and heterogeneous surfaces
Adam P. Willard and David Chandler*
Received 7th April 2008, Accepted 28th May 2008
First published as an Advance Article on the web 9th October 2008
DOI: 10.1039/b805786a
Using coarse-grained models we investigate the behavior of water adjacent to anextended hydrophobic surface peppered with various fractions of hydrophilicpatches of different sizes. We study the spatial dependence of the mean interfaceheight, the solvent density fluctuations related to drying the patchy substrate, andthe spatial dependence of interfacial fluctuations. We find that adding smalluniform attractive interactions between the substrate and solvent cause the meanposition of the interface to be very close to the substrate. Nevertheless, theinterfacial fluctuations are large and spatially heterogeneous in response to theunderlying patchy substrate. We discuss the implications of these findings forthe assembly of heterogeneous surfaces.
1 Introduction: interfaces and hydrophobicity
The interaction of liquid water with oily components in aqueous solution is centralto many phenomena.1 These phenomena are called ‘‘hydrophobic effects.’’ Thehydrophobic effects leading to robust assembly, such as micelle formation andprotein aggregation, follow from the nucleation of water/vapor-like interfaces adja-cent to sufficiently extended hydrophobic surfaces.2 This article reports on computersimulation studies of the behavior of such interfaces when the surfaces containhydrophilic regions.Our analysis focuses on probability distributions for density fluctuations in water,
distributions like Pv(N), which stands for the probability that the centers of N watermolecules are found in a volume v. We consider such distributions because arrange-ments of water molecules near hydrophobic solutes are similar to those of water nearvoids in the liquid. The connection between voids and hydrophobicity is found inStillinger’s3 proposal that a water/oil interface is similar to a water/vapor interface.This proposal is supported by recent experiments.4–6
Hummer, Pratt and their coworkers7 were the first to call attention to the behaviorof Pv(N). For small microscopic volumes in water, they used computer simulation toshow that this probability is almost exactly Gaussian. Some theories of homoge-neous liquids and hydrophobic effects are based upon the assumption thatmicroscopic density fluctuations obey Gaussian statistics,8 so the finding of ref. 7provides support for those theories. But more important is a remarkably simpleyet quantitatively accurate theory for solvation free energies of small hydrophobicspecies.7,9 This consequence follows from the fact that the solvation free energy orexcess chemical potential for a solvent excluding volume v, Dmv, is given by10
bDmv ! "ln Pv(0),
Department of Chemistry, University of California, Berkeley, California 94720, USA. E-mail:[email protected]
PAPER www.rsc.org/faraday_d | Faraday Discussions
This journal is ª The Royal Society of Chemistry 2009 Faraday Discuss., 2009, 141, 209–220 | 209
where 1/b! kBT is Boltzmann’s constant times temperature. Therefore, to the extentthat Pv(N) is Gaussian, Dmv can be expressed entirely in terms of bulk water’s meandensity and mean-square density fluctuations, both of which can be determined fromexperimentally known quantities. With this solvation free energy in hand, hydrationfree energies of real apolar molecules can be estimated, using perturbation theory toaccount for solvent–solute forces beyond those of excluded volume interactions.2,11
But for large volumes v, or for small volumes in proximity to an extended hydro-phobic surface, the Gaussian approximation to the probability Pv(N) ceases to beaccurate. Here, probabilities for fluctuations are not the same as those for thebulk liquid because the chemical potentials of liquid water and vapor differ byvery little compared to kBT. As such, a large void in water can nucleate vapor-likeconfigurations,3 and vapor-like configurations increase the likelihood of largedensity fluctuations. Thus, for large v and N much smaller than its mean, hNiv,Pv(N) is much larger than the probability predicted by the Gaussian approximation.In other words, for large enough v, Pv(N) possesses a fat non-Gaussian tail at smallvalues of N. Understanding the nature of this tail is important for estimating valuesof solvation free energies of sufficiently large clusters and extended surfaces ofhydrophobic species. It is also important for estimating likely pathways by whichwater molecules are displaced during assembly of hydrophobic clusters.A theory for the fat tail or large length scale hydrophobicity must account for the
presence of both liquid and vapor and the interface between them. The theory ofLum, Chandler andWeeks (LCW)12 does so by partitioning the density field of waterinto two components. One is the field ns(r), which describes two-phase coexistenceand is presumed to vary slowly in space (hence the subscript ‘‘s’’). The other,dr(r), is a Gaussian field, which is assumed to include all variation not capturedby ns(r). The theory predicts the onset of large length scale hydrophobicity ata size of roughly 1 nm. Variations of ns(r) occur on length scales greater than0.2 or 0.3 nm. The equations governing the behavior of ns(r) are continuum versionsof equations governing a binary field (i.e. the density field of a lattice gas or theequivalent spin field of an Ising model) on a three-dimensional cubic lattice.13,14
To the extent that a spatially coarse resolution of the liquid density is of interest,a lattice-gas field suffices to describe the fluctuations of water. The role of dr(r) inthat case is implicit, affecting the values of parameters of the lattice-gas Hamilto-nian. We have adopted this approach in earlier work modeling dynamics of a hydro-phobic chain in water,15 dynamics of water in and around nanometer-scale tubes,16
and the dynamics of dimerization for two nanometer-scale spheres in water.17 Someadditional justification for the approach can be found in theoretical analysis18 andexplicit atomistic modeling.19 Examples of other applications of the lattice-gasmodel to treat water fluctuations near extended hydrophobic surfaces are foundin ref. 20.The models we employ are defined in the next section. We then present results of
this general coarse-grained approach, examining density fluctuations adjacent toextended complex surfaces in a way that elucidates recent atomistic modeling.21,22
Some of our calculations focus directly on the fluid interface, i.e. a d! 2 dimensionalmanifold in d ! 3 dimensional space. Our principal results demonstrate the vari-ability of probabilities for dewetting fluctuations, and further show that the behav-iors of these probabilities are richer and more physically pertinent than might beguessed from the behaviors of their mean values alone.
2 Models
2.1 Lattice gas
As noted, we use the lattice-gas model as the discrete version of the slowly varyingfield of the LCW theory,12 with the parameterization of Ref. [18]. In particular thedensity field is taken as a binary field on a three dimensional cubic lattice with lattice
210 | Faraday Discuss., 2009, 141, 209–220 This journal is ª The Royal Society of Chemistry 2009
spacing l. Positions of a lattice site are specified with r ! xx + yy + zz where x, y andz are Cartesian components. In units of l these components have integer values. Thelattice site at position r has occupation number nr equal to 1 if the volume within thatsite is liquid-like and equal to 0 if the volume within that site is vapor-like. Thislattice-gas model is capable of supporting liquid/vapor phase coexistence and itsinterfaces. The energy for a given binary density field {nr} is given by,
EL#{nr}$ ! "3X
r;r0
0nrnr0 "
X
r
mnr; (1)
where the primed summation is taken over nearest-neighbor pairs of lattice sites, and
at coexistence the chemical potential, m, has value 33. To match the surface tension,
compressibility and proximity of water to coexistence, we use 3 ! 1.51kBT, l ! 0.21
nm and m ! 33 + 2.25 % 10"4kBT. In the simulation of this model, we use cells that
are periodically replicated in the x and y directions, while the upper z boundaries are
populated with liquid-like lattice sites and the lower z boundaries serve to charac-
terize patchy surfaces described below.Fluctuations across a liquid/vapor interface can be correlated over very long
distances while correlation lengths in the bulk liquid are very small. Thus, fluctua-tions of the interface can be interesting even while those in the bulk are uninteresting.For this reason we choose to simulate systems with large interfacial area (in the xyplane) and relatively little bulk liquid. Taking the perspective to the extreme woulddisregard the bulk entirely, as is done with an interface model, which we turn to now.
2.2 Fluid interface
Here we adapt the two-dimensional model of a liquid/vapor interface described byWeeks.23 In this model, a liquid/vapor interface is treated as a two-dimensionalmanifold that is coarse-grained onto a square lattice in a reference plane with latticespacing approximately equal to the bulk correlation length of water, x ! 0.42 nm.This two-dimensional interface is periodically replicated in each of the two dimen-sions of the reference xy plane. The interfacial profile is characterized by the setof height variables {ha}, where ha is the distance of the interface normal to the refer-ence xy plane, and a ! xx + yy refers to a lattice point on that plane. In this model,which we call the ‘‘fluid interface model,’’ the free interface has the energy function
EW#{ha}$ !G
2
X
a;a0
0#ha " ha0$2; (2)
where the primed summation is taken over nearest-neighbor pairs, and to be consis-
tent with the surface tension of water, we use Gx2 ! 0.1kBT.
2.3 Patchy substrate
We couple the models described above to a patchy substrate. The substrate interactswith the solvent through excluded volume interactions as well as attractive poten-tials. The substrate lies in the z ! 0 plane, and is taken to be effectively infinite inthe xy plane, and it prevents the liquid (and its interface) from existing at z < 0.In addition to excluding volume, the substrate surface contains regions which attractthe liquid or its interface. These are called ‘‘hydrophilic’’ regions. The other portionsof the substrate, which simply exclude the liquid, are the ‘‘hydrophobic’’ regions. Thesurface is partitioned with the same length scale as the solvent and the state (hydro-philic or hydrophobic) is specified with a variable sa (1 or 0 respectively). For thelattice-gas model, the interaction energy between the patchy substrate {sa} andthe first (i.e. z ! 1) layer of lattice sites is
This journal is ª The Royal Society of Chemistry 2009 Faraday Discuss., 2009, 141, 209–220 | 211
DEL#{nr}; {sa}$ ! "X
a
#3sana;1 & Dgna;1$: (3)
Here, na,z refers to nr for r! a + zz, and a specifies x and y. The strong adhesive forces
used to mimic the hydrophilic interactions come from the first term of eqn (3). The
quantity Dg is the surface adhesive interaction which is included to capture small
attractive interactions such as van der Waals attractions between oil and water. In
the absence of hydrophilic regions (sa ! 0 for all a), Dghna,1i/l 2 is the mean field esti-
mate of the difference between water/oil surface tension and water/vapor surface
tension. In this case (all sa set to zero), the choiceDgz 0.053, gives an average density
profile hna,zi consistent with that found experimentally for water/oil interfaces.4
For the fluid interface model, the interaction energy between the patchy substrate{sa} and the interfacial profile is
DEW#{ha}; {sa}$ ! W#{ha}$ "X
a
Q#hc " ha$!saEx
2 & DGx2"; (4)
where Q(x) is the Heaviside step function which is equal to 1 for x > 0 and zero
otherwise, hc is a cutoff distance equal to the bulk correlation length, the hydrophilic
interaction energy Ex2 is taken to be 3kBT (which amounts to about half the attrac-
tive energy per surface unit of 3 for the lattice gas), and the function W({ha})
excludes the interface from crossing the z ! 0 plane (it is infinite if any ha < 0,
and it is zero if all ha $ 0). The surface adhesion DG is included to capture small
attractive interactions (as above). For the case of no hydrophilic patches, the choice
of DGx2 ! 0.7kBT gives an interfacial profile which is consistent with experimental
observations for water adjacent to oil.4
2.4 Monte Carlo
For the lattice-gas model, we carry out Monte Carlo trajectories for {nr}. Accep-tance rejection obeys a detailed balance for the grand canonical ensemble with netenergy function
E({nr},{sa}) ! EL({nr}) + DEL({nr};{sa}). (5)
To specify hydrophilic patches, i.e. to specify {sa}, we tile the substrate with d %d squares thus creating a square lattice with lattice spacing d, where d is an integermultiple of l. At random, a fraction f of the d % d squares are made hydrophilic (i.e.sa ! 1 for each substrate lattice site in the square). The pattern so formed is then heldfixedwhile theMonteCarlo trajectory is carriedout for allnrvalues.Averagesover thesetrajectories are then recorded, the results of which are described in the next section.For the fluid interface model, we carry out Monte Carlo trajectories for {ha}.
Acceptance and rejection obeys a detailed balance with the net energy function
E({ha},{sa}) ! EW({ha}) + DEW({ha};{sa}). (6)
Here, the hydrophilic patterns are created as above, but now with underlying latticespacing x rather than l. Monte Carlo trajectories for ha values are performed andanalyzed with {sa} fixed. The ha values evolve continuously, unlike the nr values,which change discontinuously between 0 and 1.
3 Mean interfacial height for various substrates
We have chosen to focus on surfaces which can be characterized by two parameters.These parameters are the overall fraction f of hydrophilic sites and the size d of the
212 | Faraday Discuss., 2009, 141, 209–220 This journal is ª The Royal Society of Chemistry 2009
hydrophilic patches. The interface height of the lattice gas, ha, is defined such thatha + l is the smallest value of z, for z > 0, where na,z is not zero. For example, ifonly the cell immediately adjacent to the surface at a is empty, ha ! l. If nonewere empty, ha ! 0. In other words, ha is the value of z for the occupied lattice facetclosest to the patchy surface at a ! (x,y). Consider first the average interfacial heighthhif,d, where h/if,d denotes the equilibrium average with patchy surfaces character-ized by f and d. Specifically,
hhif ;d ! 1
Nrep
XNrep
a!1
1
Nsurf
X
a
h#a$a ; (7)
where the first summation is over Nrep different realizations of the surface patterns,
a refers to a specific realization, i.e. a ! {sa}, consistent with f and d, and the second
summation is over the Nsurf lattice sites in the patchy surface (xy plane). The quan-
tity !h(a)a is the mean height of the interface over the surface site a! (x,y), for a specific
realization
h#a$a ! 1
Nobs
XNobs
s!1
h#a$a #s$; (8)
where the summation here is over the Nobs averaging time steps in a single Monte
Carlo trajectory.In the absence of weak surface adhesive interactions, i.e. Dg ! DG ! 0, for some
fraction of hydrophilic coverages, f, the mean interface height hhif,d is non-mono-tonic in the hydrophilic patch size d. The results for these cases are shown inFig. 1. This non-monotonic behavior, which is observed in both the fluid interfacemodel as well as the lattice-gas model, demonstrates that for fixed surface composi-tion f there is an optimal patch size d for attracting the fluid.The mean interface height hhif,d is an average over a surface which is locally
heterogeneous. The influence of the local surface structure on the mean hhif,d canbe visualized for a surface realization a with the spatially resolved average interfaceheight !h(a)a . Fig. 2 shows !h(a)a for representative patchy surfaces with f! 0.03 and d! l,2l and 3l. The Figure shows that a hydrophilic patch affects interface fluctuationsbeyond the region immediately above the patch. In ref. 22, atomistic simulationswere used to study water confined between nano-scale heterogeneous surfaces.Among their findings are that the solvent density within the first hydration layerover a nano-scale hydrophobic region is increased significantly by the introductionof a border of hydrophilic sites. This effect is not symmetric with respect to theexchange of hydrophilic and hydrophobic material as the hydration layer overa nano-scale hydrophilic region is affected very little by the introduction of a hydro-phobic border. We have found (not shown here) that these results are duplicatedwith the lattice models used here.The non-monotonic behavior seen in Fig. 1 can be understood qualitatively
through the mean interfacial profile !h(a)a (shown in Fig. 2). For a surface with f !0.03 and d ! l in the absence of surface adhesive interactions the mean interfacialprofile is only slightly perturbed by the underlying hydrophilic patches. Locally,regions of the surface with a relatively high density of hydrophilic patches lowerthe average height of the interface. There are however, many isolated hydrophilicpatches for which the mean interface height shows little response. For d ! 2l,however, the hydrophilic patches pin the local interface, and because d is relativelysmall there are many such patches so that the distance between a patch and its neigh-bors is not very large. At d ! 3l there are fewer hydrophilic patches to pin the inter-face and thus the hydrophobic domains are generally larger than when d ! 2l. Theeffect of adding a small attractive interaction between the surface and solute is signif-icant as seen in Fig. 1 and 2. The mean interface is pulled much closer to the patchy
This journal is ª The Royal Society of Chemistry 2009 Faraday Discuss., 2009, 141, 209–220 | 213
surface when weak attractions are present. This result is consistent with the role ofweak attractions uncovered in earlier work.2,24–26 The non-monotonic behavior seenin the average interfacial height is eliminated and the fluctuations are reduced.
4 Density fluctuations and dewetting
To explore the onset of drying-like phenomena we have considered a finite patchysubstrate. The specific substrate size is 32 % 32 l2 for which we have studied densityfluctuations in the adjacent lattice gas. In particular, we have computed the proba-bility distribution, Pa(r) for various substrate realizations, a, corresponding todifferent hydrophilic fractions f and patch sizes d. The variable r refers to the densityin the first two solvent layers above the substrate:
Pa#r$ !#d#r" 1
2NS
X
a
'na;1 & na;2($$a; (9)
where d(/) stands for Dirac’s delta function, NS is the number of lattice sites in
a single layer of the lattice gas (in this case NS ! 322), and the averaging implied
Fig. 1 The equilibrium height, hhif,d, of the liquid vapor interface over the surface layer asa function of patch size d for the lattice-gas model (top) and the fluid interface model (bottom)at different hydrophilic fractions f. In each panel, the top three curves are computed in theabsence of weak water–substrate adhesive interactions, i.e. Dg ! DG ! 0, and the bottom threecurves are computed with weak water–substrate adhesive interactions.
214 | Faraday Discuss., 2009, 141, 209–220 This journal is ª The Royal Society of Chemistry 2009
by h/ia is carried out by umbrella sampling29 with the surface realization a fixed.
This distribution coincides with Pv(N) discussed in the introduction, in this case
where v is the combined volume of the first and second layers of cells above the
substrate with pattern a, and r multiplied by that volume is N.
4.1 Without weak surface adhesive interactions, Dg ! 0
The inset to Fig. 3 shows Pa(r) for the case of Dg ! 0 (plotted with open symbols)for representative surfaces with f ! 0.03 and d ! l, 3l and 5l. These distributions arevery broad, showing significant fluctuations over a large range of densities. Thesebroad distributions arises because the liquid/vapor interface wanders in and out ofthe observed volume. The distribution with d ! l shows an additional narrowpeak near r ! 0, which corresponds to the dewetted state where the interface hascompletely pulled away from the patchy surface.The bimodal character of Pa(r) is related to the non-monotonic behavior seen in
Fig. 1. That is, the metastable ‘‘dry’’ state can emerge when the hydrophilic patchsizes are small, but this state is absent in the case of larger patch sizes. The loss ofthe metastability occurs when hydrophilic patch sizes are large enough to pin theadjacent interface. Consider the effect of excluding volume from a fluctuatingliquid/vapor interface, as the substrate does. This imposes constraints on the config-urations accessible to the interface, and the closer the interface is to the substratethe more severe the constraints. Thus, for a purely hydrophobic substrate thereis an entropic force that drives the liquid/vapor interface away from the surface,thereby drying the substrate. By adding hydrophilic patches, the entropic drivingforce is overcome by energetically favorable interactions between the interfaceand the substrate, but just barely for the case of f ! 0.03 and d ! l, as is evidentin Fig. 3.
Fig. 2 Mean interfacial height, !h(a)a projected onto the xy plane for hydrophilic fraction f !0.03 and d! l (a), d! 2l (b), and d! 3l (c and d). The height h is indicated by the shading (scaleat right) and the location of hydrophilic surface sites (substrate sites with sa ! 1) are shownwith red circles. Panels (a), (b) and (c) correspond to averages in the absence of surface adhesiveinteractions (Dg ! 0.0) and panel (d) corresponds to an average with surface adhesive interac-tions (Dg ! 0.053). The surfaces pictured are 64 % 64 l2 in size.
This journal is ª The Royal Society of Chemistry 2009 Faraday Discuss., 2009, 141, 209–220 | 215
4.2 With weak surface adhesive interactions
The inset of Fig. 3 also shows Pa(r) for the same surface parameters, f ! 0.03 andd ! l, 3l and 5l, but with the interface adhesion turned on, i.e. Dg ! 0.053 (plottedwith filled symbols). With the interface pulled close to the surface the distributionsshift to larger values of r (more liquid-like). For d ! l the addition of adhesive inter-actions destabilizes the ‘‘dewetted’’ state and as a result the distribution is unimodaland considerably more narrow than when Dg ! 0. The main panel in Fig. 3 showsln[Pa(r)] for f ! 0.03, and d ! l, 3l and 5l. The distributions exhibit non-Gaussiantails which are a characteristic of density distributions near phase coexistence.27,28
The departure from Gaussian behavior arises because density fluctuations in thepresence of a liquid/vapor interface occur through the translation of the interface.Fluctuations without the interface require compressing a nearly incompressiblefluid. The non-Gaussian tails imply a far greater probability for large density fluctu-ations near the substrate than for those in the bulk.The quantity ln[Pa(r)] is proportional to the free energy to alter the solvent density
in the volume adjacent to the substrate, and thus reflects the cost of drying the patchysubstrate. There is a trivial dependence of the free-energy difference between ther ! 0 and r ! 1 state on the hydrophilic fraction f. For fixed f, however, the shapesof the distributions depend on the hydrophilic patch size d, and this feature is perti-nent to the kinetic pathways to drying. Moving from the wet state to the dry state asa function of r, initially near r z 0.9, a surface with larger d incurs less of a free-energy cost to dry than for a surface with a relatively small d. As r approachesr ! 0, however, for the system with smaller d, fluctuations to lower r have lowerfree energy than for systems with larger d. As the solvent density, r, over the patchysubstrate is reduced, the substrate and the hydrophilic patches dry. The solventdensity over the hydrophilic patches, the patches where sa ! 1, is given by
m1 !1
fNS
X
a
sana;1: (10)
Fig. 4 shows the average value ofm1 for substrate realization a, hm1ir,a, as a functionof r, where the average implied by h/ir,a is an average over configurations with
Fig. 3 The probability distribution for density within the first two layers of lattice sites withweak adhesive interactions (Dg ! 0.053) over the substrate surface for specific realizations ofpatchy substrates with hydrophilic fraction f ! 0.03 and patch sizes d ! l, 3l and 5l. The curvelabel ‘‘MF’’ is the result for the mean field distribution, where the substrate is uniformlyattracted to the solvent with an attractive strength equal to f3.33 The inset compares the distri-butions for d ! l (squares), d ! 3l (circles), and d ! 5l (triangles) for substrates with (hollowsymbols) and without (filled symbols) weak adhesive interactions.
216 | Faraday Discuss., 2009, 141, 209–220 This journal is ª The Royal Society of Chemistry 2009
fixed r, and substrate realization a. In Fig. 4 we see that when d ! l, the hydrophilic
patches dry essentially uniformly as the solvent density, r, is decreased. When the
hydrophilic patches are large, however, hm1ir,a does not show a significant response
to r until r is quite small, indicating that when d is large the hydrophilic patches are
among the last regions to dry. Qualitatively, therefore, the initial stages of drying are
easier for systems with larger d because there are relatively large hydrophobic
domains from which to pull the interface. The final stages of drying are difficult
when large hydrophilic patches exist, and comparatively easy for a surface with
poorly pinning small patches.
5 Spatial dependence of height fluctuations
The distribution Pa(h;a) is the probability of observing interface height h oversurface site a ! (x,y) for surface pattern a. To visualize these distributions we definethe free energy Aa(h;a),
Aa#h; a$ ! "kBT log
"Pa#h; a$
Pa
%~h#a$a ; a
&#; (11)
where h~(a)a is the most likely value of ha over the substrate at position a. Therefore,Aa(h;a) is the free energy required to move the interface at equilibrium from its most
likely position h~a to a height h. Fig. 5 shows Aa(h;a) for a surface a with f ! 0.05 and
d ! l. The weak adhesive interaction, Dg ! 0.053, causes the h ! 0 interfacial config-
uration to be favored by the interface. Fluctuations into h ! l are thermally acces-
sible to the interface (A(h;a) ) kBT) and for many regions, configurations with
h ! 2l (interface is 0.4 nm from the surface) are accessible through fluctuations of
less than 3kBT. Recall that for d ! 3l with weak adhesions turned on, the interface
is on average pulled very close to the substrate (Fig. 1), and solvent volume adjacent
to the substrate has a more liquid-like density (Fig. 3). But the distribution Aa(h;a)shows that local fluctuations in the interface through the second layer are within the
range of thermal fluctuations. On the other hand, the free energy to pull the interface
off of a hydrophilic patch is quite large. These features lead to the contrasting shapes
Fig. 4 The mean solvent density over the hydrophilic patches, hm1ir,a (see text for definition),as a function of the total solvent density in the first two solvent layers, r. Each curve is averagedover a fixed substrate realization, a, with f ! 0.03 and weak surface adhesive interactions(Dg ! 0.053).
This journal is ª The Royal Society of Chemistry 2009 Faraday Discuss., 2009, 141, 209–220 | 217
of the Pa(r) values shown in Fig. 3, as well as the response of hm1ir,a to r shown in
Fig. 4.
6 Implications
Many meso-scale solutes in nature are patchy. Hua et al.30 characterized the distri-butions of hydrophobic and hydrophilic subunits on an assembling surface ofproteins by coarse-graining the hydrophobicity on protein surfaces over 5 % 5 Asquares. The resulting hydrophobic surface distributions look like hydrophilic sitesdistributed on a hydrophobic background, not unlike the surface models we areconsidering. The relative fraction of hydrophilic sites depends upon the proteinconsidered and coincide qualitatively with our model with fz 0.25–0.50 and hydro-philic patch size d z 2l–3l.The protein surfaces studied in ref. 30 and 31 are finite in size, extending over
several nanometers. Our model surfaces are effectively infinite and thus lacka boundary term which may be significant in the actual kinetics of assembly. None-theless, for solutes with hydrophobic surfaces extending over about 1 nm, the mech-anism for assembly can involve the drying of the solvent volume between the twoassembling proteins.17,20,32 This drying event and subsequent solute aggregation ispreempted by the formation of a vapor tunnel which arises when the interfacessurrounding both solutes come close to contact. The height fluctuations of the inter-face next to these solutes describe this phenomenon. Specifically, these interfacialfluctuations set the range over which solutes can broadcast their presence into thebulk solvent. Fig. 6 shows the free energy Aa(h;a) for a substrate a with f ! 0.25and d ! 3l chosen to resemble the assembling surfaces of the proteins in ref. 30.For surfaces consistent with these parameters, the typical range which is thermallyaccessible to fluctuations of the interface height is 0–2l (0.0–0.4 nm), implying thattwo solutes with similar surface distributions in water can form a vapor tunnel at
Fig. 5 The spatial variation of the free energy, bAa(h;a) to displace the interface at a from itsmost likely height to the indicated height h for a surface realization with f ! 0.05 and with d ! land 3l in the presence of weak adhesive interactions (Dg ! 0.053). The surfaces pictured are64 % 64 l2 in size.
218 | Faraday Discuss., 2009, 141, 209–220 This journal is ª The Royal Society of Chemistry 2009
separations smaller than about 0.8 nm. Not surprisingly, therefore, Hua et al.30 findspontaneous dewetting at surface separations of 0.6 nm or less in a reasonably shortperiod of time (within 100 ps) using atomistic simuations with patchy hydrophobicprotein pairs.The expulsion of water over buried hydrophilic patches presents a barrier to the
water-mediated assembly of patchy solutes. For an isolated patchy surface, largehydrophilic patches remain ‘‘wet’’ until the final stages of drying. This is becausedrying these large patches is very costly, as manifest in the downward curving tailsfor d ! 5 in Fig. 3 near r ! 0. These downward tails do not exist for f ! 0 (notshown), which is consistent with the finding of ref. 22 that the addition of a hydro-philic site to an otherwise hydrophobic surface significantly slows drying. The freeenergy associated with fluctuations of the interface to h ! 2l (0.41 nm) over thehydrophilic patches are close to 10kBT (Fig. 6). With free-energy barriers this largeit is unlikely that the first stage of assembly occurs through a complete drying of theassembling surfaces. Thus, during assembly, water molecules over large hydrophilicregions are likely expelled in the latter stages of assembly through a mechanismdifferent from those which are responsible for drying the hydrophobic regions. Infact, retaining some amount of buried water until the final stages of assembly canbe advantageous. Specifically, water retained between assembling protein surfacescan aid in the final stages of successful aggregation.34,35 Indeed, the grouping ofhydrophilic regions can allow for much larger fluctuations of the interface overthe substrate. This is demonstrated in Fig. 3 through the comparison of the distribu-tion Pa(r) for homogeneous and patchy surfaces. Along with results for patchysubstrates, Fig. 3 has density distributions for systems with substrate layers ofuniform attractive interactions with magnitude f (these are the ‘‘mean field’’ linesin Fig. 3).33 Large fluctuations in solvent density towards the vapor-like state areless likely in the case of the homogenous substrate. We conclude that as the geometryof a patchy hydrophobic surface considerably affects the mean behavior of theadjacent solvent, it also significantly affects the interfacial fluctuations and thuscan strongly influence the kinetics of assembly.
Acknowledgements
This work was supported in its initial stages by the Director, Office of Science, Officeof Basic Energy Sciences, Chemical Sciences, Geosciences and Biosciences Division,U. S. Department of Energy under Contract No. DE-AC02-05CH11231, and thenby the National Institutes of Health. We are grateful to Lan Hua, Bruce Berneand their co-workers for sharing with us their unpublished data31 relating to thesurface distributions of self-assembling proteins.
Fig. 6 The spatial variation of the free energy bAa(h;a) for h ! 2l, and a patch pattern reali-zation with f ! 0.25 and d ! 3l. The pattern of the patchy substrate {sa} is not explicitly dis-played but is evident in the pattern of bAa(h;a). The surface pictured is 64 % 64 l2 in size.
This journal is ª The Royal Society of Chemistry 2009 Faraday Discuss., 2009, 141, 209–220 | 219
References
1 C. Tanford, The Hydrophobic Effect: Formation of Micelles and Biological Membranes,Wiley, New York, 1980.
2 D. Chandler, Nature, 2005, 437, 640.3 F. H. Stillinger, J. Solution Chem., 1973, 2, 141.4 A. Poynor, L. Hong, I. K. Robinson, S. Granick, Z. Zhang and P. A. Fenter, Phys. Rev.Lett., 2006, 97, 266101.
5 M. Maccarini, R. Steitz, M. Himmelhaus, J. Fick, S. Tatur, M. Wolff, M. Grunze,J. Janecek and R. R. Netz, Langmuir, 2007, 23, 598.
6 M. Mezger, H. Reichert, S. Schoder, J. Okasinski, H. Schroder, H. Dosch, D. Palms,J. Ralston and V. Honkimaki, Proc. Natl. Acad. Sci. U. S. A., 2006, 103, 18401.
7 G. Hummer, S. Garde, A. Pohorille, A. E. Garcia and L. Pratt, Proc. Natl. Acad. Sci.U. S. A., 1996, 93, 8951.
8 D. Chandler, Phys. Rev. E, 1993, 48, 2898.9 S. Garde, G. Hummer, A. E. Garcia, M. E. Paulaitis and L. R. Pratt, Phys. Rev. Lett., 1996,77, 4966.
10 B. Widom, J. Chem. Phys., 1963, 39, 2808.11 L. R. Pratt and D. Chandler, J. Chem. Phys., 1977, 67, 3683.12 K. Lum, D. Chandler and J. D. Weeks, J. Phys. Chem. B, 1999, 103, 4570.13 D. Chandler, Introduction to Modern Statistical Mechanics, Oxford University Press, New
York, 1987.14 P. M. Chaikin and T. C. Lubensky, Principals of Condensed Matter Physics, Cambridge
University Press, Cambridge, 2000.15 P. R. ten Wolde and D. Chandler, Proc. Natl. Acad. Sci. U. S. A., 2002, 99, 6539.16 L. Maibaum, A. R. Dinner and D. Chandler, J. Phys. Chem. B, 2004, 108, 6778.17 A. P. Willard and D. Chandler, J. Phys. Chem. B, 2008, 112, 6187.18 P. R. ten Wolde, S. X. Sun and D. Chandler, Phys. Rev. E, 2001, 65, 011201.19 T. F.Miller, E. Vanden-Eijnden and D. Chandler, Proc. Natl. Acad. Sci. U. S. A., 2007, 104,
14559.20 For example: A. Luzar and K. Leung, J. Chem. Phys., 2000, 113, 5836; K. Leung and
A. Luzar, J. Chem. Phys., 2000, 113, 5845; T. A. McCormick, Phys. Rev. E, 2003, 68,061601; K. Lum and A. Luzar, Phys. Rev. E, 1997, 56, R6283.
21 N. Giovambattista, C. F. Lopez, P. J. Rossky and P. G. Debenedetti, Proc. Natl. Acad. Sci.U. S. A., 2008, 105, 2274.
22 N. Giovambattista, P. J. Rossky and P. G. Debenedetti, J. Phys. Chem. C, 2007, 111, 1323.23 J. D. Weeks, J. Chem. Phys., 1977, 67, 3106.24 L. Maibaum and D. Chandler, J. Phys. Chem. B, 2007, 111, 9025.25 D. M. Huang and D. Chandler, J. Phys. Chem. B, 2002, 106, 2047.26 H. S. Ashbaugh and M. E. Paulaitis, J. Am. Chem. Soc., 2001, 123, 10721.27 S. T. Bramwell, J.-Y. Fortin, P. C. W. Holdsworth, S. Peysson, J. F. Pinton, B. Portelli and
M. Sellitto, Phys. Rev. E, 2001, 63, 041106.28 D. M. Huang and D. Chandler, Phys. Rev. E, 2000, 61, 1501.29 D. Frenkel, B. Smit, Understanding Molecular Simulation, Academic Press, London, 2nd
edn, 2002.30 L. Hua, X. H. Huang, P. Liu, R. H. Zhou and B. J. Berne, J. Phys. Chem. B, 2007, 111,
9069.31 L. Hua, private communication, 2008.32 X. Huang, R. Zhou and B. J. Berne, J. Phys. Chem. B, 2005, 109, 3546.33 The uniform mean field substrate with hydrophilic fraction f interacts with the adjacent
lattice gas with the energy DE#MF$L #{nr}$ ! "
Pa#f 3& Dg$na;1.
34 Y. Levy and J. N. Onuchic, Proc. Natl. Acad. Sci. U. S. A., 2004, 101, 3325.35 F. B. Sheinerman and C. L. Brooks III, J. Mol. Biol., 1998, 278, 439.
220 | Faraday Discuss., 2009, 141, 209–220 This journal is ª The Royal Society of Chemistry 2009